Fifteen years in engineering in the UK, nearly a decade of it in the car industry, and I could not have told you what an API was. Now I run a bespoke software and automation agency out of the UK, with clients here and in the United States. This is the unglamorous record of what happened in between — plus straight answers to the questions businesses actually ask me about AI, taken from the calls where they asked them.
An engineering owner asked me this last week, almost word for word. The answer is yes, and the reason it works has very little to do with the AI.
Somebody asked me where all this ends. Here is the line I draw, and the test I use to decide which side of it a job sits on.
A publisher said something to me last week that explains more failed AI projects than any report I have read. He knows owners who would not connect their inbox to one of these tools whatever it did for them, and the reason was not the technology.
We got the whole team onto a call and watched them log in. It went well. Then the thing sat untouched for nearly three weeks.
A sales manager told me the application in front of him would take sixty hours. Then he said the thing I have repeated to three people since.
This question decides whether a project is straightforward or a nightmare, and it nearly always gets asked last. Two companies asked me it this month and I couldn't answer either of them.
Someone asked me where my time goes and I could not answer it with anything other than a feeling. His prescription was to log everything for thirty days, business and personal. The only reason I agreed is that I am not the one doing the logging.
A director asked me this about an online shop that has been working quietly for four years. It is the most reasonable objection there is, and the answer has almost nothing to do with features.
One question came up on my calls more than any other this summer, and it was a small one: is it finished yet? Here is what I changed so the answer arrives sooner.
Websites have come up on most of my calls lately. The request usually starts with how the site looks. The conversation that follows rarely stays there.
When owners picture a new system, they picture what they'll see on screen. The years of records sitting in the old one are what most often catch them out.
The most useful thing a client sent me this month was a screen recording of somebody doing a dull, repetitive task. It changed what I proposed, and it cost her about ten minutes.
If you run the UK arm of a bigger group, the decision about new software often isn't yours. Here is what I would do before anyone writes a proposal.
My old plan for growth was more clients, more hours and more of me. I've capped it instead, and this is the arithmetic that changed my mind.
The old one was made with an AI site builder. The new one is plain files I own. Both choices were right, at different times.
A short list for the version of me who had just decided to do this and had no idea what he was walking into.
Four requests I hear constantly, and what the person is usually describing underneath.
It started as hypocrisy avoidance. It became my best sales asset.
My proposals ran past twenty pages. Nobody read past four.
Six questions that separate people who have delivered from people who have read about it.
Yes — and here is what genuinely differs, plus the practical things that bite.
It is not the technical test. It is what happens when I describe something badly.
Including the test that matters most, which is not a technical one.
Everything shipped. Everything worked. Adoption was one person. The most useful failure I have had.
Clients sometimes want to go faster and bigger. Here is why I hold the line.
A client asked me this directly, which was a good sign about them and a hard question about me.
They have not changed in two years and they have never once failed to find the real problem.
Yes, without drama. Here is what I will and will not agree to, and why.
I kept a tally of every feature I talked a client out of. The number surprised me.
Two deals moved, one client paused, and I learned what my pipeline actually was.
A question from the real world that a lot of software ignores, at its cost.
The number that matters is not headcount or turnover.
Four questions on the back of an envelope, and the trap that catches almost everyone.
Yes, and it is often the best arrangement — provided one thing is settled up front.
One of the most common ways good software quietly dies, and it is preventable.
Adoption is the whole game, and it is decided long before the training session.
A checklist you can hold any supplier to, including me.
The money was good and the project was interesting. Everything else was wrong, and I nearly ignored it.
Fixed price per phase, and the reason is about who carries which risk.
This is the healthiest possible starting position, and I would rather have it than a detailed spec.
The single most common worry, and the answer is more reassuring than most people expect — with one condition.
Everything I had been warned about was wrong, and the thing nobody mentioned nearly sank it.
You will, and that is normal. What matters is that it is visible rather than absorbed in silence.
Good. That should be possible on any Tuesday, without a negotiation.
You do, in an account with your name on it, and I will explain exactly why I insist on that.
Doing this alongside something else had a ceiling, and I hit it. What actually changed was not the hours.
Yes, and it is more common than anyone admits. Here is what I look at before saying yes.
It matters commercially, because one of them is much cheaper and more reliable than the other.
Agree the measure before the build, because afterwards everyone remembers the baseline differently.
A fair challenge: you have not built anything yet. Here is exactly what you get and why it is not a sales document.
Less than you would think, and the reason why is the useful part.
I opened every meeting undermining myself. It was costing me work and it was not even accurate.
Usually alongside them, not instead of them — and getting this wrong makes your life much harder.
You do not need to read their code. You need to tell whether what they built matches what the client meant.
Asked often, usually as a proxy for "do you know what you are doing". Here is a straight answer.
The word does two jobs at once and you do not get to choose which one it does in the other person.
Yes. In fact I will usually push you towards it even when you are ready to spend more.
Underestimated by nearly every supplier, including me, early on. Here are the real numbers.
I underpriced a build so badly that finishing it properly cost me money. What I learned was not about numbers.
The question that should be asked before you sign and usually is not. One client admitted he never had that conversation, and it cost him.
"Will the AI integrate into existing software, or will you create a separate application?" The answer decides whether anyone uses it.
Eight words that turn a vague automation conversation into a decision.
Sometimes you should, and I will tell you when. Here is where it stops working.
Asked on more calls than any other compliance question. Not legal advice — just what I have actually had to get right.
Free scoping produced polite meetings and vague proposals. Charging changed the work, the access, and the answer.
Asked directly, and it deserved a better answer than the one I gave at the time.
"We gave it a go-ahead to discovery and write some software — how long does that take?" A straight answer, including the bit that is your side, not mine.
A client asked for eleven features. I delivered eleven features. Two got used, and that was my fault, not theirs.
"Why don't I own the report?" was asked of someone else, in front of me, and it changed how I write contracts.
Scoping and coaching are the same job in different clothes: getting the real problem out of someone who has stopped noticing it.
"Did this hallucinate or what?" It will, sometimes. The question is what you built around it.
Every enquiry that opens with "we need an AI strategy" is a person describing a Thursday they never get back.
Rarely asked out loud by the person paying. Always present in the room. Here is how I handle it.
"Why are we spending so much money on software when we could have something off the shelf for a few hundred quid a month?" A completely reasonable question with a real answer.
It was small, it was unglamorous, and I have never been more frightened of an invoice.
A prospect asked me this twice in one meeting, which told me I had not answered it properly the first time. Here is the answer I give now.
The honest answer to the question everyone wants answered on the first call, including why I will not give you a number in the first ten minutes.
The most technical thing I did in a week was reschedule a Zoom. This is where it started.